Bu kaydın yasal hükümlere uygun olmadığını düşünüyorsanız lütfen sayfa sonundaki Hata Bildir bağlantısını takip ederek bildirimde bulununuz. Kayıtlar ilgili üniversite yöneticileri tarafından eklenmektedir. Nadiren de olsa kayıtlarla ilgili hatalar oluşabilmektedir. MİTOS internet üzerindeki herhangi bir ödev sitesi değildir!

A SYSTEM PROPOSAL FOR COUNTING THE NUMBER OF PEOPLE IN STILL IMAGES

Oluşturulma Tarihi: 08-02-2017

Niteleme Bilgileri

Tür: Tez

Alt Tür: Yüksek Lisans Tezi

Yayınlanma Durumu: Yayınlanmamış

Dosya Biçimi: PDF

Dil: İngilizce

Konu(lar): Teknoloji (Genel),

Yazar(lar): Al Zubaidi, Waleed Khalid Mahmood (Yazar),

Emeği Geçen(ler): Koyuncu, Murat (Danışman),

Anahtar Kelimeler

Counting people in still images, face detection, human body detection, head detection, Viola-Jones algorithm, HOG, crowded images, morphological image operation, circular Hough transform.


Özet

Counting people in images is a challenging task in the computer vison. This thesis targets to estimate the number of people in images accurately. Our focus is on implementing a robust counting algorithm that depends on different approaches together to detect humans having different appearances in images. Three different approaches are taken as the base of the proposed people counting method. These approaches are frontal face detection, human whole body detection, and people head detection.The main contribution of this thesis is using different approaches together for counting people in still images. We have done that by using Viola-Jones algorithm for face detection, the HOG features and SVM classifier for human body detection, and Hough transform and morphological image processing for head detection in crowded images. Any image is processed by three detectors in parallel and detected people are counted. Then, their results are combined for a final decision. The proposed method is implemented in the C++ language with the help of the OpenCV image processing library. The proposed method is tested and compared with some other approaches. The experimental results show that the proposed method achieves more successful results compared to other approaches when test dataset includes various image categories.


İçindekiler



Açıklamalar



Haklar



Notlar



Kaynakça


Atıf Yapanlar

Gözat Sayfasına Dön

 

Sosyal Medya ve Araçlar

İstatistikler

  • Kayıt
    • Bu ay: 4
    • Toplam: 2441
  • Online
    • Ziyaretçi: 26
    • Üye: 0
    • Toplam: 26

Detaylı İstatistikler